Dataset Health Analyzer

ReadyHow to use

Score CSV health across missing values, duplicates, row shape, and type consistency.

AnalysisSummaryBrowser workspace

Free to use. No account, no daily limit, and no backend processing for tool input.

Tip: use samples, upload, copy, download, and send-to actions inside the workspace where available.

What is Dataset Health Analyzer?

Dataset Health Analyzer profiles a CSV and scores it across the four things that actually break downstream work: missing values, duplicate rows, ragged row shape, and columns whose type is inconsistent from row to row. It gives you a single health score to compare datasets against each other, and a per-column breakdown showing which columns are dragging that score down and by how much — so cleaning starts with the column that matters rather than the first one in the file.

How to use Dataset Health Analyzer

1Drop in a CSV, or paste a sample of it.
2Read the overall health score as a quick comparison figure.
3Work down the per-column breakdown, which is ordered by how much each column costs you.
4Check the row-shape report for rows with too few or too many fields.
5Take the flagged columns into the matching cleaning tool.

Features

Single health score for comparing datasets at a glance
Per-column missing-value counts and percentages
Duplicate row detection with counts
Ragged row detection — rows whose field count differs from the header
Type-consistency checking per column
Runs in the browser on a bounded preview, so large files stay responsive

Common uses

Deciding whether a dataset is usable before building anything on it
Triaging which columns to clean first
Comparing the quality of two candidate data sources
Producing a defensible quality summary for a report

Used by

Data analysts
Researchers
Business analysts
Students
Analysts

Common errors

Treating a high score as clean — the score covers structural quality, not whether the values are correct or the data is current.
Reading empty strings and nulls as the same thing; they often mean different things ("not answered" versus "answered as blank") and the report distinguishes them.
Assuming duplicate rows are always errors — in event and transaction data, identical rows can be entirely legitimate.
Judging a whole file from the preview when the problems cluster near the end, which is common in appended exports.

How it works

1The file is parsed with the header taken as the expected shape, so any row with a different field count is flagged immediately.
2Each column is scanned for empty values, and for the set of types its values parse as — a column that is 98% integers and 2% text is a type-consistency problem worth knowing about.
3The component scores are combined into a single figure, with each column's contribution kept separate so the breakdown stays actionable.

Recommended workflow

1
Fix what was found
Apply the common repairs in one pass.
Open tool
2
Look at missing data in detail
Per-column missing-value report with patterns.
Open tool
3
Investigate the duplicates
See which rows repeat and how often.
Open tool

Frequently asked questions

What counts as a duplicate row?: A row whose every field matches another row exactly. Near-duplicates — same record with a differently formatted phone number — are not caught here; that is what fuzzy matching is for.
Is my data uploaded to be analysed?: No. Parsing and scoring happen in your browser. The site is a static export with no backend, so there is nowhere for the file to be sent.
What is a good score?: Compare rather than target. The score is most useful for ranking datasets against each other and for tracking one dataset across cleaning passes, not as an absolute pass mark.
Why does it analyse only a preview of large files?: To keep the browser responsive and within memory. The preview cap is shown alongside the results, so you always know how much of the file the numbers describe.